gdr — Gemini Deep Research CLI

Status: Alpha. APIs and commands may change before v1.0.

Disclaimer: Unofficial, community-built CLI. Not affiliated with, endorsed by, or sponsored by Google LLC. "Gemini" and "Deep Research" are trademarks of Google LLC, used here nominatively to describe the APIs this tool interacts with.

A terminal-first client for Google's Gemini Deep Research and Deep Research Max agents. Run long-horizon research tasks from your shell and get cleanly organized artifacts — markdown reports, charts, citations, and a full transcript — saved to disk.

gdr is a thin, honest wrapper over the google-genai SDK. The SDK does the work; gdr adds ergonomics: streaming UI, safe config/secret management, local history, a collaborative planning flow, resume-after-disconnect, and safe MCP wiring.

Why

Deep Research tasks run for 5–60 minutes. Running them from a web UI means keeping a browser tab open. gdr decouples the task from your terminal session: start a research run, live-stream thought summaries, walk away, resume later by ID, and get artifacts on disk that fit into any downstream pipeline.

Install

# With pipx (recommended — isolated install)
pipx install gemini-deep-research

# With uv tool
uv tool install gemini-deep-research

# From source (dev)
git clone https://github.com/johnswyou/gemini-deep-research
cd gemini-deep-research
uv sync --extra dev
uv run gdr --help

Quickstart

export GEMINI_API_KEY=...             # get one at https://aistudio.google.com/apikey

# One-line research with the fast agent
gdr research "Latest trends in RISC-V adoption"

# Maximum-quality agent for due diligence
gdr research --max "Competitive landscape of EV batteries"

# Review and refine the agent's plan before it spends tokens
gdr research --plan "Impact of AI on semiconductor supply chain"

# Ground in your own documents
gdr research --file ~/Downloads/10k.pdf \
  "Compare risk factors vs our 2024 filing"

Each run produces a timestamped directory under ~/gdr-reports/ containing report.md, sources.json, transcript.json, metadata.json, and any generated images.

Command reference

Run gdr --help or gdr <command> --help for full flag reference.

See docs/USAGE.md for long-form command documentation, docs/MCP.md for MCP server integration, and docs/TROUBLESHOOTING.md for common failure modes and recovery recipes. Runnable demos live in examples/.

Configuration

Config lives at ~/.config/gdr/config.toml. Run gdr doctor --fix to scaffold it. Environment variables referenced with env:VAR_NAME are expanded at load time so secrets stay out of the file.

api_key = "env:GEMINI_API_KEY"
default_agent = "deep-research-preview-04-2026"
output_dir = "~/gdr-reports"
auto_open = true
confirm_max = true                  # prompt before running the Max agent
default_tools = ["google_search", "url_context", "code_execution"]
thinking_summaries = "auto"         # "auto" or "none"
visualization = "auto"              # "auto" or "off"
safe_untrusted = true               # auto-strip dangerous tools when --file/--url is used

[mcp_servers.factset]
url = "https://mcp.factset.com"
headers.Authorization = "Bearer env:FACTSET_TOKEN"

Safety

Deep Research agents can read files and the public web. gdr ships with:

• Redaction of MCP auth headers and API keys from transcript.json and from --dry-run previews (--reveal opts out).
• Path confinement: derived artifact directories are confined to the configured output_dir and slug names are sanitized (an explicit --output is honored verbatim).
• Header validation for MCP servers (no CRLF injection, no reserved names).
• --untrusted-input flag that disables code_execution and mcp_server tools for a run — use when grounding in attacker-controlled files or URLs.

See docs/MCP.md for the MCP security model and docs/TROUBLESHOOTING.md for the --untrusted-input recipe.

Development

uv sync --extra dev
uv run ruff check .
uv run mypy src
uv run pytest -q

Roadmap

Deferred to v1.1: HTML/PDF export, cost estimation, SQLite history backend, interactive setup wizard. See docs/USAGE.md for the currently-shipping surface.

License

MIT. See LICENSE.

Credits

Built on Google's Gemini Interactions API and the google-genai Python SDK.